plugin-discovery

name	plugin-discovery
description	This skill should be used when the user asks to \\\\\\\"推荐插件\\\\\\\", \\\\\\\"查找插件\\\\\\\", \\\\\\\"搜索插件\\\\\\\", \\\\\\\"有什么插件\\\\\\\", \\\\\\\"安装插件\\\\\\\", \\\\\\\"plugin recommendation\\\\\\\", \\\\\\\"find plugins\\\\\\\", \\\\\\\"search plugins\\\\\\\", or needs help discovering and managing CodeBuddy Code plugins from marketplaces.
description_zh	智能插件发现和管理助手 - 支持智能搜索、多插件并行对比、多插件协同工作流（sequence-run）、插件信息详解、许愿新插件等功能
version	0.1.0
source	codebuddy
source_plugin	plugin-finder

Plugin Discovery and Management

Purpose

This skill provides comprehensive guidance for discovering, recommending, installing, and comparing plugins from AgentZ Code and AgentZ plugin marketplaces. Enable intelligent plugin discovery through AI semantic matching, streamlined installation workflows, and multi-plugin comparison capabilities.

When to Use

Activate this skill when users:

• Request plugin recommendations for specific needs
• Want to explore available plugins in marketplaces
• Need help installing plugins
• Want to compare multiple plugins for the same task
• Ask general questions about plugin capabilities

Core Concepts

Dual Platform Support

AgentZ Code supports both AgentZ and Claude plugin ecosystems:

AgentZ Platform:

• Marketplace registry: ~/.AgentZ/plugins/known_marketplaces.json
• Plugin installation uses AgentZ-specific paths
• Uses .AgentZ-plugin/ directory structure

Claude Platform:

• Marketplace registry: ~/.claude/plugins/known_marketplaces.json
• Installed plugins: ~/.claude/plugins/installed_plugins.json
• Uses .claude-plugin/ directory structure

Marketplace Structure

Each marketplace in known_marketplaces.json contains:

• name: Marketplace identifier
• installLocation: Local path to marketplace
• manifest: Full marketplace.json content with plugin listings
• lastUpdated: Last update timestamp

Example entry:

{
  "claude-plugins-official": {
    "type": "git",
    "source": {
      "source": "git",
      "url": "https://github.com/anthropics/claude-plugins-official"
    },
    "installLocation": "/Users/user/.AgentZ/plugins/marketplaces/...",
    "manifest": {
      "plugins": [...]
    }
  }
}

Plugin Metadata

Each plugin in marketplace.json includes:

• name: Plugin identifier
• description: What the plugin does
• category: Plugin type (development, productivity, security, etc.)
• keywords: Search tags
• tags: Additional metadata
• source: Plugin location (local path or git URL)
• version: Plugin version

Plugin Discovery Workflow

Step 1: Read Marketplace Data

Read marketplace registry files to get all available plugins:

# AgentZ marketplaces
cat ~/.AgentZ/plugins/known_marketplaces.json

# Claude marketplaces
cat ~/.claude/plugins/known_marketplaces.json

Extract plugin information from the manifest.plugins array in each marketplace.

Step 2: Semantic Matching

Use AI to understand user intent and match against plugin metadata:

Match against:

1. Description - Primary matching field, use semantic similarity
2. Keywords - Exact and partial matches
3. Category - Broad classification matching
4. Tags - Additional metadata matching
5. Name - Fuzzy name matching

Scoring algorithm:

• Description semantic match: 50 points
• Keywords exact match: 30 points
• Category match: 15 points
• Tags match: 5 points

Recommend plugins scoring above 40 points.

Step 3: Present Recommendations

Use chat_ui tool tool to display recommendations with:

• Plugin name and marketplace
• Description
• Category and keywords
• Installation count (if available)
• Multi-select option enabled

Example presentation:

Question: "Which plugins would you like to install?"
Options:
1. code-review@claude-plugins-official
   Description: Automated code review with multiple specialized agents
   Category: productivity
   
2. security-guidance@claude-plugins-official
   Description: Security reminder hook for potential security issues
   Category: security

Step 4: Install Selected Plugins

For each selected plugin, execute installation command:

/plugin install plugin-name@marketplace-name

Batch installation:

/plugin install plugin1@market1 plugin2@market2 plugin3@market3

Step 5: Post-Installation

After installation, always inform user:

Critical reminder:

✅ Plugins installed successfully!

⚠️  IMPORTANT: You must restart AgentZ Code for plugins to take effect.

Steps:
1. Exit AgentZ Code completely
2. Relaunch AgentZ Code
3. Verify plugins loaded with /plugin list

Plugin Comparison Workflow

When comparing multiple plugins for the same task:

Step 1: Identify Relevant Plugins

Search installed plugins that could handle the task:

# List installed plugins
/plugin list

Match task requirements against:

• Plugin description
• Plugin category
• Known capabilities from plugin.json

Step 2: Determine Invocation Method

For each relevant plugin, check what components it provides:

Check plugin structure:

ls ~/.AgentZ/plugins/cache/marketplace-name/plugin-name/

Invocation priority:

1. If plugin has agents → Invoke agent with task description
2. If plugin has commands → Execute relevant command
3. If plugin has skills → Load skill and use in current context
4. If plugin has hooks → Note: hooks run automatically, not invoked

Step 3: Parallel Execution

Execute each plugin's capability concurrently:

# Create parallel tasks
Task 1: Use plugin-1's agent
Task 2: Use plugin-2's command
Task 3: Apply plugin-3's skill

Track outputs and artifacts from each execution.

Step 4: Results Collection

Collect for each plugin:

• Output quality: Completeness, accuracy, usefulness
• Execution time: How long it took
• Artifacts generated: Files, reports, modifications
• User experience: Ease of use, clarity

Step 5: AI Analysis and Comparison

Analyze collected results across dimensions:

Quality comparison:

• Which plugin produced most accurate results?
• Which had most comprehensive coverage?
• Which provided most actionable insights?

Performance comparison:

• Which was fastest?
• Which used resources most efficiently?

Output comparison:

• Which generated most useful artifacts?
• Which had best formatting and presentation?

Overall assessment:

• Recommend best plugin for this specific task
• Note strengths of each plugin
• Suggest when to use alternatives

Generate comparison report:

## Plugin Comparison Report

### Task: [Task Description]

### Plugins Tested
1. plugin-1@marketplace-1
2. plugin-2@marketplace-2
3. plugin-3@marketplace-3

### Results Summary

**plugin-1:**
- Quality: ⭐⭐⭐⭐⭐ (5/5)
- Speed: ⭐⭐⭐⭐ (4/5)
- Output: [Description of output]
- Strengths: [List strengths]
- Weaknesses: [List weaknesses]

[Repeat for other plugins]

### Recommendation

For this specific task, **plugin-1** is recommended because:
- [Reason 1]
- [Reason 2]

However, consider **plugin-2** when:
- [Scenario 1]
- [Scenario 2]

### Detailed Comparison

[Detailed analysis of differences]

Installation Commands Reference

Basic Installation

/plugin install plugin-name@marketplace-name

Batch Installation

/plugin install plugin1@market1 plugin2@market2

List Installed Plugins

/plugin list

Uninstall Plugin

/plugin uninstall plugin-name

Update Marketplace

/plugin marketplace update marketplace-name

User Configuration

Users can configure plugin finder behavior via ~/.AgentZ/.local.md:

---
auto_recommend: true
recommendation_threshold: medium
show_install_count: true
preferred_marketplaces:
  - AgentZ-plugins-official
  - claude-plugins-official
---

Configuration options:

• auto_recommend: Enable automatic plugin recommendations (default: true)
• recommendation_threshold: Matching threshold - high/medium/low (default: medium)
• show_install_count: Display installation popularity (default: true)
• preferred_marketplaces: Priority order for searching (default: all)

Reading configuration:

# Check if config exists
if [ -f ~/.AgentZ/.local.md ]; then
  # Parse YAML frontmatter
  # Apply settings
fi

Best Practices

Discovery

1. Search all marketplaces: Don't limit to single marketplace unless user specifies
2. Show diverse results: Include plugins from different categories if relevant
3. Explain recommendations: Always provide reasoning for each recommendation
4. Respect user preferences: Honor settings from .local.md configuration

Installation

1. Verify marketplace exists: Check plugin is in known marketplaces
2. Batch when possible: Install multiple plugins in single command
3. Always remind to restart: Critical for plugins to take effect
4. Confirm success: Verify installation completed

Comparison

1. Clear task definition: Ensure task is well-defined before comparing
2. Fair testing: Use same inputs for all plugins
3. Multiple dimensions: Evaluate quality, speed, output, UX
4. Actionable recommendations: Clear guidance on which to use when

Troubleshooting

Plugin Not Found

Error: Plugin 'name' not found in any marketplace

Solutions:

1. Update marketplaces: /plugin marketplace update
2. Verify plugin name spelling
3. Check if plugin was removed from marketplace

Installation Fails

Error: Failed to install plugin-name@marketplace

Solutions:

1. Check internet connectivity (for remote sources)
2. Verify marketplace location exists
3. Check disk space
4. Review error message details

Plugin Not Loading After Install

Issue: Plugin installed but not showing in /plugin list

Solutions:

1. Primary: Restart AgentZ Code (exit completely and relaunch)
2. Verify installation: ls ~/.AgentZ/plugins/cache/
3. Check plugin.json syntax
4. Review AgentZ Code logs

Additional Resources

Reference Files

For detailed implementation patterns:

• references/marketplace-format.md - Complete marketplace.json schema
• references/matching-algorithm.md - Semantic matching implementation details

Scripts

Utility scripts in scripts/:

• search-plugins.sh - Search across all marketplaces
• compare-plugins.sh - Compare plugin capabilities

Examples

Working examples in examples/:

• search-example.md - Example search workflow
• install-example.md - Example installation process
• compare-example.md - Example comparison report

Implementation Notes

When implementing plugin finder commands and agents:

1. Read both platform registries: Check both ~/.AgentZ/ and ~/.claude/ paths
2. Handle missing files gracefully: Marketplace files may not exist
3. Parse JSON carefully: Marketplace structure varies by version
4. Use semantic matching: Don't rely solely on keyword matching
5. Provide multi-select UI: Allow users to install multiple plugins at once
6. Track execution context: Remember what plugins are being compared
7. Generate comprehensive reports: Make comparison results actionable

Success Criteria

A successful plugin discovery interaction:

✅ User describes need clearly or vaguely ✅ AI understands intent through semantic matching ✅ Relevant plugins identified from all marketplaces ✅ Recommendations presented with clear reasoning ✅ User selects plugins easily (multi-select) ✅ Installation completes successfully ✅ User reminded to restart ✅ Plugins load correctly after restart

A successful plugin comparison:

✅ Task clearly defined ✅ Relevant installed plugins identified ✅ All plugins tested fairly with same inputs ✅ Results collected across multiple dimensions ✅ AI analysis provides clear insights ✅ Recommendation actionable and justified ✅ Report helps user make informed decision